Add-on: your own infrastructure only
Globally Certified AI Penetration Testing
An autonomous agent tests your external attack surface, then proves what it finds.
Your own infrastructure. Nothing else.
This can only ever run against systems you own, or that you are contractually authorised to test. If the target is not yours and you cannot prove permission, we will not test it. There is no version of this engagement that starts without that proof.
Every other Kizuna service is passive. It reads public and leaked data and never sends a single request to the subject. This one is the exception: it actively engages the target. That is why it is sold separately, why it can only ever run against infrastructure you own or are contractually authorised to test, and why a person on our team releases every engagement by hand.
How you buy it
On this site, priced on how many assets you want tested. Paying opens the engagement; it does not start it.
What it costs
Priced per engagement. Scope decides the number, so the scope conversation comes first.
When it starts
When a human on our team releases it. Never automatically, and never from a payment.
What you get
Findings that are proven, not guessed at.
- External attack surface enumerated and mapped
- Real vulnerabilities probed and safely verified, not a list of maybes
- Evidence for every finding, so you can reproduce it
- Remediation guidance, prioritised
- A full audit log of every action the agent took
What is actually behind it
Not a chatbot with a prompt pointed at your servers.
That is the fair assumption to make about anything calling itself AI penetration testing, so it is worth being specific. This is a set of purpose-built agents drawing on more than one model, standing on years of real engagement data, and corrected every day by the testers using it on live work.
Several models, not one
Different stages of a test need different reasoning, so the agents draw on more than one LLM rather than routing everything through a single general-purpose one.
Years of engagement data
Custom agents built on years of real testing, not a prompt written against public documentation. That history is what tells them which leads are worth pulling.
Improved daily, by testers
It is in live use every day, and qualified penetration testers feed it every day: what it found, what it walked past, what a human would have tried next. It is retrained on that rather than left to go stale.
And no, not your engagement
The fair next question is whether your test becomes someone else's training data. Anything that feeds the agents is obfuscated first, there is no route from what they learn back to the engagement it came from, and none of it is exposed to any public platform.
The authorisation gate
Four steps. All four, in this order, before anything is touched.
This is the product working as designed. Skipping a step is not something we can be talked into.
- 1
Prove the assets are yours
A DNS TXT record or a hosted verification file on each target domain, plus evidence of company ownership. That means incorporation documents, or a signed letter of authority if you are testing on a client's behalf.
- 2
Sign the rules of engagement
Scope, the exact target domains and IPs, permitted test windows, and explicit written permission to test.
- 3
List what is in scope
Everything is out of scope by default. Only the externally-facing assets you list explicitly can be touched.
- 4
A human releases the run
Nothing dispatches from a checkout. One of our operators reviews the authorisation pack and starts the engagement.
Guardrails
We will not DoS you, phish your staff, or pivot into your network.
An autonomous agent with a free hand is a liability, so this one does not have a free hand. These are the limits it runs under, and they are a feature of the engagement rather than a disclaimer at the bottom of it.
- External attack surface only. No internal or lateral movement, no pivoting into your network
- No social engineering or phishing of your staff
- No denial-of-service or stress testing
- No destructive actions, and no data taken beyond the minimum needed to prove a finding
- Agreed test windows and rate limits honoured; the run aborts on contact with anything out of scope
Why the gate exists
Active security testing against infrastructure you do not own or are not authorised to test is a criminal offence in most jurisdictions, including the UK Computer Misuse Act, the US Computer Fraud and Abuse Act, and UAE cybercrime law. The authorisation gate is not a formality.
Scope it and order it
You choose how many assets are in scope and list exactly which ones. Paying opens the engagement and puts it in front of an operator: the four steps above happen before an agent is pointed at anything, and if we cannot verify your authorisation we refund it rather than test.
What the order form asks for
- Your company or legal entity
- The externally-facing domains and IPs you want in scope
- Whether you own those assets, or hold a signed letter of authority
- A preferred test window, and anything that must stay out of scope
Prefer to write it yourself? Email search@kizuna-ai.io.